HAMMER 28/many: Implement zoned blockmap
[dragonfly.git] / sys / vfs / hammer / hammer_vnops.c
1 /*
2  * Copyright (c) 2007 The DragonFly Project.  All rights reserved.
3  * 
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  * 
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  * 
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  * 
34  * $DragonFly: src/sys/vfs/hammer/hammer_vnops.c,v 1.30 2008/02/10 09:51:01 dillon Exp $
35  */
36
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/fcntl.h>
41 #include <sys/namecache.h>
42 #include <sys/vnode.h>
43 #include <sys/lockf.h>
44 #include <sys/event.h>
45 #include <sys/stat.h>
46 #include <sys/dirent.h>
47 #include <vm/vm_extern.h>
48 #include <vfs/fifofs/fifo.h>
49 #include "hammer.h"
50
51 /*
52  * USERFS VNOPS
53  */
54 /*static int hammer_vop_vnoperate(struct vop_generic_args *);*/
55 static int hammer_vop_fsync(struct vop_fsync_args *);
56 static int hammer_vop_read(struct vop_read_args *);
57 static int hammer_vop_write(struct vop_write_args *);
58 static int hammer_vop_access(struct vop_access_args *);
59 static int hammer_vop_advlock(struct vop_advlock_args *);
60 static int hammer_vop_close(struct vop_close_args *);
61 static int hammer_vop_ncreate(struct vop_ncreate_args *);
62 static int hammer_vop_getattr(struct vop_getattr_args *);
63 static int hammer_vop_nresolve(struct vop_nresolve_args *);
64 static int hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
65 static int hammer_vop_nlink(struct vop_nlink_args *);
66 static int hammer_vop_nmkdir(struct vop_nmkdir_args *);
67 static int hammer_vop_nmknod(struct vop_nmknod_args *);
68 static int hammer_vop_open(struct vop_open_args *);
69 static int hammer_vop_pathconf(struct vop_pathconf_args *);
70 static int hammer_vop_print(struct vop_print_args *);
71 static int hammer_vop_readdir(struct vop_readdir_args *);
72 static int hammer_vop_readlink(struct vop_readlink_args *);
73 static int hammer_vop_nremove(struct vop_nremove_args *);
74 static int hammer_vop_nrename(struct vop_nrename_args *);
75 static int hammer_vop_nrmdir(struct vop_nrmdir_args *);
76 static int hammer_vop_setattr(struct vop_setattr_args *);
77 static int hammer_vop_strategy(struct vop_strategy_args *);
78 static int hammer_vop_nsymlink(struct vop_nsymlink_args *);
79 static int hammer_vop_nwhiteout(struct vop_nwhiteout_args *);
80 static int hammer_vop_ioctl(struct vop_ioctl_args *);
81 static int hammer_vop_mountctl(struct vop_mountctl_args *);
82
83 static int hammer_vop_fifoclose (struct vop_close_args *);
84 static int hammer_vop_fiforead (struct vop_read_args *);
85 static int hammer_vop_fifowrite (struct vop_write_args *);
86
87 static int hammer_vop_specclose (struct vop_close_args *);
88 static int hammer_vop_specread (struct vop_read_args *);
89 static int hammer_vop_specwrite (struct vop_write_args *);
90
91 struct vop_ops hammer_vnode_vops = {
92         .vop_default =          vop_defaultop,
93         .vop_fsync =            hammer_vop_fsync,
94         .vop_getpages =         vop_stdgetpages,
95         .vop_putpages =         vop_stdputpages,
96         .vop_read =             hammer_vop_read,
97         .vop_write =            hammer_vop_write,
98         .vop_access =           hammer_vop_access,
99         .vop_advlock =          hammer_vop_advlock,
100         .vop_close =            hammer_vop_close,
101         .vop_ncreate =          hammer_vop_ncreate,
102         .vop_getattr =          hammer_vop_getattr,
103         .vop_inactive =         hammer_vop_inactive,
104         .vop_reclaim =          hammer_vop_reclaim,
105         .vop_nresolve =         hammer_vop_nresolve,
106         .vop_nlookupdotdot =    hammer_vop_nlookupdotdot,
107         .vop_nlink =            hammer_vop_nlink,
108         .vop_nmkdir =           hammer_vop_nmkdir,
109         .vop_nmknod =           hammer_vop_nmknod,
110         .vop_open =             hammer_vop_open,
111         .vop_pathconf =         hammer_vop_pathconf,
112         .vop_print =            hammer_vop_print,
113         .vop_readdir =          hammer_vop_readdir,
114         .vop_readlink =         hammer_vop_readlink,
115         .vop_nremove =          hammer_vop_nremove,
116         .vop_nrename =          hammer_vop_nrename,
117         .vop_nrmdir =           hammer_vop_nrmdir,
118         .vop_setattr =          hammer_vop_setattr,
119         .vop_strategy =         hammer_vop_strategy,
120         .vop_nsymlink =         hammer_vop_nsymlink,
121         .vop_nwhiteout =        hammer_vop_nwhiteout,
122         .vop_ioctl =            hammer_vop_ioctl,
123         .vop_mountctl =         hammer_vop_mountctl
124 };
125
126 struct vop_ops hammer_spec_vops = {
127         .vop_default =          spec_vnoperate,
128         .vop_fsync =            hammer_vop_fsync,
129         .vop_read =             hammer_vop_specread,
130         .vop_write =            hammer_vop_specwrite,
131         .vop_access =           hammer_vop_access,
132         .vop_close =            hammer_vop_specclose,
133         .vop_getattr =          hammer_vop_getattr,
134         .vop_inactive =         hammer_vop_inactive,
135         .vop_reclaim =          hammer_vop_reclaim,
136         .vop_setattr =          hammer_vop_setattr
137 };
138
139 struct vop_ops hammer_fifo_vops = {
140         .vop_default =          fifo_vnoperate,
141         .vop_fsync =            hammer_vop_fsync,
142         .vop_read =             hammer_vop_fiforead,
143         .vop_write =            hammer_vop_fifowrite,
144         .vop_access =           hammer_vop_access,
145         .vop_close =            hammer_vop_fifoclose,
146         .vop_getattr =          hammer_vop_getattr,
147         .vop_inactive =         hammer_vop_inactive,
148         .vop_reclaim =          hammer_vop_reclaim,
149         .vop_setattr =          hammer_vop_setattr
150 };
151
152 static int hammer_dounlink(struct nchandle *nch, struct vnode *dvp,
153                            struct ucred *cred, int flags);
154 static int hammer_vop_strategy_read(struct vop_strategy_args *ap);
155 static int hammer_vop_strategy_write(struct vop_strategy_args *ap);
156
157 #if 0
158 static
159 int
160 hammer_vop_vnoperate(struct vop_generic_args *)
161 {
162         return (VOCALL(&hammer_vnode_vops, ap));
163 }
164 #endif
165
166 /*
167  * hammer_vop_fsync { vp, waitfor }
168  */
169 static
170 int
171 hammer_vop_fsync(struct vop_fsync_args *ap)
172 {
173         hammer_inode_t ip;
174         int error;
175
176         ip = VTOI(ap->a_vp);
177         error = hammer_sync_inode(ip, ap->a_waitfor, 0);
178         return (error);
179 }
180
181 /*
182  * hammer_vop_read { vp, uio, ioflag, cred }
183  */
184 static
185 int
186 hammer_vop_read(struct vop_read_args *ap)
187 {
188         struct hammer_transaction trans;
189         hammer_inode_t ip;
190         off_t offset;
191         struct buf *bp;
192         struct uio *uio;
193         int error;
194         int n;
195         int seqcount;
196
197         if (ap->a_vp->v_type != VREG)
198                 return (EINVAL);
199         ip = VTOI(ap->a_vp);
200         error = 0;
201         seqcount = ap->a_ioflag >> 16;
202
203         hammer_start_transaction(&trans, ip->hmp);
204
205         /*
206          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
207          */
208         uio = ap->a_uio;
209         while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_rec.ino_size) {
210                 offset = uio->uio_offset & HAMMER_BUFMASK;
211 #if 0
212                 error = cluster_read(ap->a_vp, ip->ino_rec.ino_size,
213                                      uio->uio_offset - offset, HAMMER_BUFSIZE,
214                                      MAXBSIZE, seqcount, &bp);
215 #endif
216                 error = bread(ap->a_vp, uio->uio_offset - offset,
217                               HAMMER_BUFSIZE, &bp);
218                 if (error) {
219                         brelse(bp);
220                         break;
221                 }
222                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
223                 n = HAMMER_BUFSIZE - offset;
224                 if (n > uio->uio_resid)
225                         n = uio->uio_resid;
226                 if (n > ip->ino_rec.ino_size - uio->uio_offset)
227                         n = (int)(ip->ino_rec.ino_size - uio->uio_offset);
228                 error = uiomove((char *)bp->b_data + offset, n, uio);
229                 if (error) {
230                         bqrelse(bp);
231                         break;
232                 }
233                 if ((ip->flags & HAMMER_INODE_RO) == 0 &&
234                     (ip->hmp->mp->mnt_flag & MNT_NOATIME) == 0) {
235                         ip->ino_rec.ino_atime = trans.tid;
236                         hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
237                 }
238                 bqrelse(bp);
239         }
240         hammer_commit_transaction(&trans);
241         return (error);
242 }
243
244 /*
245  * hammer_vop_write { vp, uio, ioflag, cred }
246  */
247 static
248 int
249 hammer_vop_write(struct vop_write_args *ap)
250 {
251         struct hammer_transaction trans;
252         struct hammer_inode *ip;
253         struct uio *uio;
254         off_t offset;
255         struct buf *bp;
256         int error;
257         int n;
258         int flags;
259
260         if (ap->a_vp->v_type != VREG)
261                 return (EINVAL);
262         ip = VTOI(ap->a_vp);
263         error = 0;
264
265         if (ip->flags & HAMMER_INODE_RO)
266                 return (EROFS);
267
268         /*
269          * Create a transaction to cover the operations we perform.
270          */
271         hammer_start_transaction(&trans, ip->hmp);
272         uio = ap->a_uio;
273
274         /*
275          * Check append mode
276          */
277         if (ap->a_ioflag & IO_APPEND)
278                 uio->uio_offset = ip->ino_rec.ino_size;
279
280         /*
281          * Check for illegal write offsets.  Valid range is 0...2^63-1
282          */
283         if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0) {
284                 hammer_commit_transaction(&trans);
285                 return (EFBIG);
286         }
287
288         /*
289          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
290          */
291         while (uio->uio_resid > 0) {
292                 int fixsize = 0;
293
294                 offset = uio->uio_offset & HAMMER_BUFMASK;
295                 n = HAMMER_BUFSIZE - offset;
296                 if (n > uio->uio_resid)
297                         n = uio->uio_resid;
298                 if (uio->uio_offset + n > ip->ino_rec.ino_size) {
299                         vnode_pager_setsize(ap->a_vp, uio->uio_offset + n);
300                         fixsize = 1;
301                 }
302
303                 if (uio->uio_segflg == UIO_NOCOPY) {
304                         /*
305                          * Issuing a write with the same data backing the
306                          * buffer.  Instantiate the buffer to collect the
307                          * backing vm pages, then read-in any missing bits.
308                          *
309                          * This case is used by vop_stdputpages().
310                          */
311                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
312                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
313                         if ((bp->b_flags & B_CACHE) == 0) {
314                                 bqrelse(bp);
315                                 error = bread(ap->a_vp,
316                                               uio->uio_offset - offset,
317                                               HAMMER_BUFSIZE, &bp);
318                         }
319                 } else if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
320                         /*
321                          * entirely overwrite the buffer
322                          */
323                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
324                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
325                 } else if (offset == 0 && uio->uio_offset >= ip->ino_rec.ino_size) {
326                         /*
327                          * XXX
328                          */
329                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
330                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
331                         vfs_bio_clrbuf(bp);
332                 } else {
333                         /*
334                          * Partial overwrite, read in any missing bits then
335                          * replace the portion being written.
336                          */
337                         error = bread(ap->a_vp, uio->uio_offset - offset,
338                                       HAMMER_BUFSIZE, &bp);
339                         if (error == 0)
340                                 bheavy(bp);
341                 }
342                 if (error == 0)
343                         error = uiomove((char *)bp->b_data + offset, n, uio);
344
345                 /*
346                  * If we screwed up we have to undo any VM size changes we
347                  * made.
348                  */
349                 if (error) {
350                         brelse(bp);
351                         if (fixsize) {
352                                 vtruncbuf(ap->a_vp, ip->ino_rec.ino_size,
353                                           HAMMER_BUFSIZE);
354                         }
355                         break;
356                 }
357                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
358                 if (ip->ino_rec.ino_size < uio->uio_offset) {
359                         ip->ino_rec.ino_size = uio->uio_offset;
360                         flags = HAMMER_INODE_RDIRTY;
361                         vnode_pager_setsize(ap->a_vp, ip->ino_rec.ino_size);
362                 } else {
363                         flags = 0;
364                 }
365                 ip->ino_rec.ino_mtime = trans.tid;
366                 flags |= HAMMER_INODE_ITIMES | HAMMER_INODE_BUFS;
367                 hammer_modify_inode(&trans, ip, flags);
368
369                 /*
370                  * The file write must be tagged with the same TID as the
371                  * inode, for consistency in case the inode changed size.
372                  * This guarantees the on-disk data records will have a
373                  * TID <= the inode TID representing the size change.
374                  *
375                  * If a prior write has not yet flushed, retain its TID.
376                  */
377                 if (bp->b_tid == 0)
378                         bp->b_tid = ip->last_tid;
379
380                 if (ap->a_ioflag & IO_SYNC) {
381                         bwrite(bp);
382                 } else if (ap->a_ioflag & IO_DIRECT) {
383                         bawrite(bp);
384                 } else {
385                         bdwrite(bp);
386                 }
387         }
388         if (error)
389                 hammer_abort_transaction(&trans);
390         else
391                 hammer_commit_transaction(&trans);
392         return (error);
393 }
394
395 /*
396  * hammer_vop_access { vp, mode, cred }
397  */
398 static
399 int
400 hammer_vop_access(struct vop_access_args *ap)
401 {
402         struct hammer_inode *ip = VTOI(ap->a_vp);
403         uid_t uid;
404         gid_t gid;
405         int error;
406
407         uid = hammer_to_unix_xid(&ip->ino_data.uid);
408         gid = hammer_to_unix_xid(&ip->ino_data.gid);
409
410         error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
411                                   ip->ino_data.uflags);
412         return (error);
413 }
414
415 /*
416  * hammer_vop_advlock { vp, id, op, fl, flags }
417  */
418 static
419 int
420 hammer_vop_advlock(struct vop_advlock_args *ap)
421 {
422         struct hammer_inode *ip = VTOI(ap->a_vp);
423
424         return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size));
425 }
426
427 /*
428  * hammer_vop_close { vp, fflag }
429  */
430 static
431 int
432 hammer_vop_close(struct vop_close_args *ap)
433 {
434         return (vop_stdclose(ap));
435 }
436
437 /*
438  * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
439  *
440  * The operating system has already ensured that the directory entry
441  * does not exist and done all appropriate namespace locking.
442  */
443 static
444 int
445 hammer_vop_ncreate(struct vop_ncreate_args *ap)
446 {
447         struct hammer_transaction trans;
448         struct hammer_inode *dip;
449         struct hammer_inode *nip;
450         struct nchandle *nch;
451         int error;
452
453         nch = ap->a_nch;
454         dip = VTOI(ap->a_dvp);
455
456         if (dip->flags & HAMMER_INODE_RO)
457                 return (EROFS);
458
459         /*
460          * Create a transaction to cover the operations we perform.
461          */
462         hammer_start_transaction(&trans, dip->hmp);
463
464         /*
465          * Create a new filesystem object of the requested type.  The
466          * returned inode will be referenced but not locked.
467          */
468
469         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
470         if (error)
471                 kprintf("hammer_create_inode error %d\n", error);
472         if (error) {
473                 hammer_abort_transaction(&trans);
474                 *ap->a_vpp = NULL;
475                 return (error);
476         }
477
478         /*
479          * Add the new filesystem object to the directory.  This will also
480          * bump the inode's link count.
481          */
482         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
483         if (error)
484                 kprintf("hammer_ip_add_directory error %d\n", error);
485
486         /*
487          * Finish up.
488          */
489         if (error) {
490                 hammer_rel_inode(nip, 0);
491                 hammer_abort_transaction(&trans);
492                 *ap->a_vpp = NULL;
493         } else {
494                 hammer_commit_transaction(&trans);
495                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
496                 hammer_rel_inode(nip, 0);
497                 if (error == 0) {
498                         cache_setunresolved(ap->a_nch);
499                         cache_setvp(ap->a_nch, *ap->a_vpp);
500                 }
501         }
502         return (error);
503 }
504
505 /*
506  * hammer_vop_getattr { vp, vap }
507  */
508 static
509 int
510 hammer_vop_getattr(struct vop_getattr_args *ap)
511 {
512         struct hammer_inode *ip = VTOI(ap->a_vp);
513         struct vattr *vap = ap->a_vap;
514
515 #if 0
516         if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
517             (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
518             ip->obj_asof == XXX
519         ) {
520                 /* LAZYMOD XXX */
521         }
522         hammer_itimes(ap->a_vp);
523 #endif
524
525         vap->va_fsid = ip->hmp->fsid_udev;
526         vap->va_fileid = ip->ino_rec.base.base.obj_id;
527         vap->va_mode = ip->ino_data.mode;
528         vap->va_nlink = ip->ino_rec.ino_nlinks;
529         vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
530         vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
531         vap->va_rmajor = 0;
532         vap->va_rminor = 0;
533         vap->va_size = ip->ino_rec.ino_size;
534         hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime);
535         hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime);
536         hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
537         vap->va_flags = ip->ino_data.uflags;
538         vap->va_gen = 1;        /* hammer inums are unique for all time */
539         vap->va_blocksize = 32768; /* XXX - extract from root volume */
540         vap->va_bytes = ip->ino_rec.ino_size;
541         vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type);
542         vap->va_filerev = 0;    /* XXX */
543         /* mtime uniquely identifies any adjustments made to the file */
544         vap->va_fsmid = ip->ino_rec.ino_mtime;
545         vap->va_uid_uuid = ip->ino_data.uid;
546         vap->va_gid_uuid = ip->ino_data.gid;
547         vap->va_fsid_uuid = ip->hmp->fsid;
548         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
549                           VA_FSID_UUID_VALID;
550
551         switch (ip->ino_rec.base.base.obj_type) {
552         case HAMMER_OBJTYPE_CDEV:
553         case HAMMER_OBJTYPE_BDEV:
554                 vap->va_rmajor = ip->ino_data.rmajor;
555                 vap->va_rminor = ip->ino_data.rminor;
556                 break;
557         default:
558                 break;
559         }
560
561         return(0);
562 }
563
564 /*
565  * hammer_vop_nresolve { nch, dvp, cred }
566  *
567  * Locate the requested directory entry.
568  */
569 static
570 int
571 hammer_vop_nresolve(struct vop_nresolve_args *ap)
572 {
573         struct namecache *ncp;
574         hammer_inode_t dip;
575         hammer_inode_t ip;
576         hammer_tid_t asof;
577         struct hammer_cursor cursor;
578         union hammer_record_ondisk *rec;
579         struct vnode *vp;
580         int64_t namekey;
581         int error;
582         int i;
583         int nlen;
584         int flags;
585         u_int64_t obj_id;
586
587         /*
588          * Misc initialization, plus handle as-of name extensions.  Look for
589          * the '@@' extension.  Note that as-of files and directories cannot
590          * be modified.
591          */
592         dip = VTOI(ap->a_dvp);
593         ncp = ap->a_nch->ncp;
594         asof = dip->obj_asof;
595         nlen = ncp->nc_nlen;
596         flags = dip->flags;
597
598         for (i = 0; i < nlen; ++i) {
599                 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
600                         asof = hammer_str_to_tid(ncp->nc_name + i + 2);
601                         flags |= HAMMER_INODE_RO;
602                         break;
603                 }
604         }
605         nlen = i;
606
607         /*
608          * If there is no path component the time extension is relative to
609          * dip.
610          */
611         if (nlen == 0) {
612                 ip = hammer_get_inode(dip->hmp, &dip->cache[1], dip->obj_id,
613                                       asof, flags, &error);
614                 if (error == 0) {
615                         error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
616                         hammer_rel_inode(ip, 0);
617                 } else {
618                         vp = NULL;
619                 }
620                 if (error == 0) {
621                         vn_unlock(vp);
622                         cache_setvp(ap->a_nch, vp);
623                         vrele(vp);
624                 }
625                 return(error);
626         }
627
628         /*
629          * Calculate the namekey and setup the key range for the scan.  This
630          * works kinda like a chained hash table where the lower 32 bits
631          * of the namekey synthesize the chain.
632          *
633          * The key range is inclusive of both key_beg and key_end.
634          */
635         namekey = hammer_directory_namekey(ncp->nc_name, nlen);
636
637         error = hammer_init_cursor_hmp(&cursor, &dip->cache[0], dip->hmp);
638         cursor.key_beg.obj_id = dip->obj_id;
639         cursor.key_beg.key = namekey;
640         cursor.key_beg.create_tid = 0;
641         cursor.key_beg.delete_tid = 0;
642         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
643         cursor.key_beg.obj_type = 0;
644
645         cursor.key_end = cursor.key_beg;
646         cursor.key_end.key |= 0xFFFFFFFFULL;
647         cursor.asof = asof;
648         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
649
650         /*
651          * Scan all matching records (the chain), locate the one matching
652          * the requested path component.
653          *
654          * The hammer_ip_*() functions merge in-memory records with on-disk
655          * records for the purposes of the search.
656          */
657         if (error == 0)
658                 error = hammer_ip_first(&cursor, dip);
659
660         rec = NULL;
661         obj_id = 0;
662
663         while (error == 0) {
664                 error = hammer_ip_resolve_data(&cursor);
665                 if (error)
666                         break;
667                 rec = cursor.record;
668                 if (nlen == rec->entry.base.data_len &&
669                     bcmp(ncp->nc_name, cursor.data, nlen) == 0) {
670                         obj_id = rec->entry.obj_id;
671                         break;
672                 }
673                 error = hammer_ip_next(&cursor);
674         }
675         hammer_done_cursor(&cursor);
676         if (error == 0) {
677                 ip = hammer_get_inode(dip->hmp, &dip->cache[1],
678                                       obj_id, asof, flags, &error);
679                 if (error == 0) {
680                         error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
681                         hammer_rel_inode(ip, 0);
682                 } else {
683                         vp = NULL;
684                 }
685                 if (error == 0) {
686                         vn_unlock(vp);
687                         cache_setvp(ap->a_nch, vp);
688                         vrele(vp);
689                 }
690         } else if (error == ENOENT) {
691                 cache_setvp(ap->a_nch, NULL);
692         }
693         return (error);
694 }
695
696 /*
697  * hammer_vop_nlookupdotdot { dvp, vpp, cred }
698  *
699  * Locate the parent directory of a directory vnode.
700  *
701  * dvp is referenced but not locked.  *vpp must be returned referenced and
702  * locked.  A parent_obj_id of 0 does not necessarily indicate that we are
703  * at the root, instead it could indicate that the directory we were in was
704  * removed.
705  *
706  * NOTE: as-of sequences are not linked into the directory structure.  If
707  * we are at the root with a different asof then the mount point, reload
708  * the same directory with the mount point's asof.   I'm not sure what this
709  * will do to NFS.  We encode ASOF stamps in NFS file handles so it might not
710  * get confused, but it hasn't been tested.
711  */
712 static
713 int
714 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
715 {
716         struct hammer_inode *dip;
717         struct hammer_inode *ip;
718         int64_t parent_obj_id;
719         hammer_tid_t asof;
720         int error;
721
722         dip = VTOI(ap->a_dvp);
723         asof = dip->obj_asof;
724         parent_obj_id = dip->ino_data.parent_obj_id;
725
726         if (parent_obj_id == 0) {
727                 if (dip->obj_id == HAMMER_OBJID_ROOT &&
728                    asof != dip->hmp->asof) {
729                         parent_obj_id = dip->obj_id;
730                         asof = dip->hmp->asof;
731                         *ap->a_fakename = kmalloc(19, M_TEMP, M_WAITOK);
732                         ksnprintf(*ap->a_fakename, 19, "0x%016llx",
733                                    dip->obj_asof);
734                 } else {
735                         *ap->a_vpp = NULL;
736                         return ENOENT;
737                 }
738         }
739
740         ip = hammer_get_inode(dip->hmp, &dip->cache[1], parent_obj_id,
741                               asof, dip->flags, &error);
742         if (ip == NULL) {
743                 *ap->a_vpp = NULL;
744                 return(error);
745         }
746         error = hammer_get_vnode(ip, LK_EXCLUSIVE, ap->a_vpp);
747         hammer_rel_inode(ip, 0);
748         return (error);
749 }
750
751 /*
752  * hammer_vop_nlink { nch, dvp, vp, cred }
753  */
754 static
755 int
756 hammer_vop_nlink(struct vop_nlink_args *ap)
757 {
758         struct hammer_transaction trans;
759         struct hammer_inode *dip;
760         struct hammer_inode *ip;
761         struct nchandle *nch;
762         int error;
763
764         nch = ap->a_nch;
765         dip = VTOI(ap->a_dvp);
766         ip = VTOI(ap->a_vp);
767
768         if (dip->flags & HAMMER_INODE_RO)
769                 return (EROFS);
770         if (ip->flags & HAMMER_INODE_RO)
771                 return (EROFS);
772
773         /*
774          * Create a transaction to cover the operations we perform.
775          */
776         hammer_start_transaction(&trans, dip->hmp);
777
778         /*
779          * Add the filesystem object to the directory.  Note that neither
780          * dip nor ip are referenced or locked, but their vnodes are
781          * referenced.  This function will bump the inode's link count.
782          */
783         error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
784
785         /*
786          * Finish up.
787          */
788         if (error) {
789                 hammer_abort_transaction(&trans);
790         } else {
791                 cache_setunresolved(nch);
792                 cache_setvp(nch, ap->a_vp);
793                 hammer_commit_transaction(&trans);
794         }
795         return (error);
796 }
797
798 /*
799  * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
800  *
801  * The operating system has already ensured that the directory entry
802  * does not exist and done all appropriate namespace locking.
803  */
804 static
805 int
806 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
807 {
808         struct hammer_transaction trans;
809         struct hammer_inode *dip;
810         struct hammer_inode *nip;
811         struct nchandle *nch;
812         int error;
813
814         nch = ap->a_nch;
815         dip = VTOI(ap->a_dvp);
816
817         if (dip->flags & HAMMER_INODE_RO)
818                 return (EROFS);
819
820         /*
821          * Create a transaction to cover the operations we perform.
822          */
823         hammer_start_transaction(&trans, dip->hmp);
824
825         /*
826          * Create a new filesystem object of the requested type.  The
827          * returned inode will be referenced but not locked.
828          */
829         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
830         if (error)
831                 kprintf("hammer_mkdir error %d\n", error);
832         if (error) {
833                 hammer_abort_transaction(&trans);
834                 *ap->a_vpp = NULL;
835                 return (error);
836         }
837
838         /*
839          * Add the new filesystem object to the directory.  This will also
840          * bump the inode's link count.
841          */
842         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
843         if (error)
844                 kprintf("hammer_mkdir (add) error %d\n", error);
845
846         /*
847          * Finish up.
848          */
849         if (error) {
850                 hammer_rel_inode(nip, 0);
851                 hammer_abort_transaction(&trans);
852                 *ap->a_vpp = NULL;
853         } else {
854                 hammer_commit_transaction(&trans);
855                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
856                 hammer_rel_inode(nip, 0);
857                 if (error == 0) {
858                         cache_setunresolved(ap->a_nch);
859                         cache_setvp(ap->a_nch, *ap->a_vpp);
860                 }
861         }
862         return (error);
863 }
864
865 /*
866  * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
867  *
868  * The operating system has already ensured that the directory entry
869  * does not exist and done all appropriate namespace locking.
870  */
871 static
872 int
873 hammer_vop_nmknod(struct vop_nmknod_args *ap)
874 {
875         struct hammer_transaction trans;
876         struct hammer_inode *dip;
877         struct hammer_inode *nip;
878         struct nchandle *nch;
879         int error;
880
881         nch = ap->a_nch;
882         dip = VTOI(ap->a_dvp);
883
884         if (dip->flags & HAMMER_INODE_RO)
885                 return (EROFS);
886
887         /*
888          * Create a transaction to cover the operations we perform.
889          */
890         hammer_start_transaction(&trans, dip->hmp);
891
892         /*
893          * Create a new filesystem object of the requested type.  The
894          * returned inode will be referenced but not locked.
895          */
896         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
897         if (error) {
898                 hammer_abort_transaction(&trans);
899                 *ap->a_vpp = NULL;
900                 return (error);
901         }
902
903         /*
904          * Add the new filesystem object to the directory.  This will also
905          * bump the inode's link count.
906          */
907         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
908
909         /*
910          * Finish up.
911          */
912         if (error) {
913                 hammer_rel_inode(nip, 0);
914                 hammer_abort_transaction(&trans);
915                 *ap->a_vpp = NULL;
916         } else {
917                 hammer_commit_transaction(&trans);
918                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
919                 hammer_rel_inode(nip, 0);
920                 if (error == 0) {
921                         cache_setunresolved(ap->a_nch);
922                         cache_setvp(ap->a_nch, *ap->a_vpp);
923                 }
924         }
925         return (error);
926 }
927
928 /*
929  * hammer_vop_open { vp, mode, cred, fp }
930  */
931 static
932 int
933 hammer_vop_open(struct vop_open_args *ap)
934 {
935         if ((ap->a_mode & FWRITE) && (VTOI(ap->a_vp)->flags & HAMMER_INODE_RO))
936                 return (EROFS);
937
938         return(vop_stdopen(ap));
939 }
940
941 /*
942  * hammer_vop_pathconf { vp, name, retval }
943  */
944 static
945 int
946 hammer_vop_pathconf(struct vop_pathconf_args *ap)
947 {
948         return EOPNOTSUPP;
949 }
950
951 /*
952  * hammer_vop_print { vp }
953  */
954 static
955 int
956 hammer_vop_print(struct vop_print_args *ap)
957 {
958         return EOPNOTSUPP;
959 }
960
961 /*
962  * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
963  */
964 static
965 int
966 hammer_vop_readdir(struct vop_readdir_args *ap)
967 {
968         struct hammer_cursor cursor;
969         struct hammer_inode *ip;
970         struct uio *uio;
971         hammer_record_ondisk_t rec;
972         hammer_base_elm_t base;
973         int error;
974         int cookie_index;
975         int ncookies;
976         off_t *cookies;
977         off_t saveoff;
978         int r;
979
980         ip = VTOI(ap->a_vp);
981         uio = ap->a_uio;
982         saveoff = uio->uio_offset;
983
984         if (ap->a_ncookies) {
985                 ncookies = uio->uio_resid / 16 + 1;
986                 if (ncookies > 1024)
987                         ncookies = 1024;
988                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
989                 cookie_index = 0;
990         } else {
991                 ncookies = -1;
992                 cookies = NULL;
993                 cookie_index = 0;
994         }
995
996         /*
997          * Handle artificial entries
998          */
999         error = 0;
1000         if (saveoff == 0) {
1001                 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
1002                 if (r)
1003                         goto done;
1004                 if (cookies)
1005                         cookies[cookie_index] = saveoff;
1006                 ++saveoff;
1007                 ++cookie_index;
1008                 if (cookie_index == ncookies)
1009                         goto done;
1010         }
1011         if (saveoff == 1) {
1012                 if (ip->ino_data.parent_obj_id) {
1013                         r = vop_write_dirent(&error, uio,
1014                                              ip->ino_data.parent_obj_id,
1015                                              DT_DIR, 2, "..");
1016                 } else {
1017                         r = vop_write_dirent(&error, uio,
1018                                              ip->obj_id, DT_DIR, 2, "..");
1019                 }
1020                 if (r)
1021                         goto done;
1022                 if (cookies)
1023                         cookies[cookie_index] = saveoff;
1024                 ++saveoff;
1025                 ++cookie_index;
1026                 if (cookie_index == ncookies)
1027                         goto done;
1028         }
1029
1030         /*
1031          * Key range (begin and end inclusive) to scan.  Directory keys
1032          * directly translate to a 64 bit 'seek' position.
1033          */
1034         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1035         cursor.key_beg.obj_id = ip->obj_id;
1036         cursor.key_beg.create_tid = 0;
1037         cursor.key_beg.delete_tid = 0;
1038         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1039         cursor.key_beg.obj_type = 0;
1040         cursor.key_beg.key = saveoff;
1041
1042         cursor.key_end = cursor.key_beg;
1043         cursor.key_end.key = HAMMER_MAX_KEY;
1044         cursor.asof = ip->obj_asof;
1045         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1046
1047         error = hammer_ip_first(&cursor, ip);
1048
1049         while (error == 0) {
1050                 error = hammer_ip_resolve_record_and_data(&cursor);
1051                 if (error)
1052                         break;
1053                 rec = cursor.record;
1054                 base = &rec->base.base;
1055                 saveoff = base->key;
1056
1057                 if (base->obj_id != ip->obj_id)
1058                         panic("readdir: bad record at %p", cursor.node);
1059
1060                 r = vop_write_dirent(
1061                              &error, uio, rec->entry.obj_id,
1062                              hammer_get_dtype(rec->entry.base.base.obj_type),
1063                              rec->entry.base.data_len,
1064                              (void *)cursor.data);
1065                 if (r)
1066                         break;
1067                 ++saveoff;
1068                 if (cookies)
1069                         cookies[cookie_index] = base->key;
1070                 ++cookie_index;
1071                 if (cookie_index == ncookies)
1072                         break;
1073                 error = hammer_ip_next(&cursor);
1074         }
1075         hammer_done_cursor(&cursor);
1076
1077 done:
1078         if (ap->a_eofflag)
1079                 *ap->a_eofflag = (error == ENOENT);
1080         uio->uio_offset = saveoff;
1081         if (error && cookie_index == 0) {
1082                 if (error == ENOENT)
1083                         error = 0;
1084                 if (cookies) {
1085                         kfree(cookies, M_TEMP);
1086                         *ap->a_ncookies = 0;
1087                         *ap->a_cookies = NULL;
1088                 }
1089         } else {
1090                 if (error == ENOENT)
1091                         error = 0;
1092                 if (cookies) {
1093                         *ap->a_ncookies = cookie_index;
1094                         *ap->a_cookies = cookies;
1095                 }
1096         }
1097         return(error);
1098 }
1099
1100 /*
1101  * hammer_vop_readlink { vp, uio, cred }
1102  */
1103 static
1104 int
1105 hammer_vop_readlink(struct vop_readlink_args *ap)
1106 {
1107         struct hammer_cursor cursor;
1108         struct hammer_inode *ip;
1109         int error;
1110
1111         ip = VTOI(ap->a_vp);
1112         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1113
1114         /*
1115          * Key range (begin and end inclusive) to scan.  Directory keys
1116          * directly translate to a 64 bit 'seek' position.
1117          */
1118         cursor.key_beg.obj_id = ip->obj_id;
1119         cursor.key_beg.create_tid = 0;
1120         cursor.key_beg.delete_tid = 0;
1121         cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1122         cursor.key_beg.obj_type = 0;
1123         cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1124         cursor.asof = ip->obj_asof;
1125         cursor.flags |= HAMMER_CURSOR_ASOF;
1126
1127         error = hammer_ip_lookup(&cursor, ip);
1128         if (error == 0) {
1129                 error = hammer_ip_resolve_data(&cursor);
1130                 if (error == 0) {
1131                         error = uiomove((char *)cursor.data,
1132                                         cursor.record->base.data_len,
1133                                         ap->a_uio);
1134                 }
1135         }
1136         hammer_done_cursor(&cursor);
1137         return(error);
1138 }
1139
1140 /*
1141  * hammer_vop_nremove { nch, dvp, cred }
1142  */
1143 static
1144 int
1145 hammer_vop_nremove(struct vop_nremove_args *ap)
1146 {
1147         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1148 }
1149
1150 /*
1151  * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1152  */
1153 static
1154 int
1155 hammer_vop_nrename(struct vop_nrename_args *ap)
1156 {
1157         struct hammer_transaction trans;
1158         struct namecache *fncp;
1159         struct namecache *tncp;
1160         struct hammer_inode *fdip;
1161         struct hammer_inode *tdip;
1162         struct hammer_inode *ip;
1163         struct hammer_cursor cursor;
1164         union hammer_record_ondisk *rec;
1165         int64_t namekey;
1166         int error;
1167
1168         fdip = VTOI(ap->a_fdvp);
1169         tdip = VTOI(ap->a_tdvp);
1170         fncp = ap->a_fnch->ncp;
1171         tncp = ap->a_tnch->ncp;
1172         ip = VTOI(fncp->nc_vp);
1173         KKASSERT(ip != NULL);
1174
1175         if (fdip->flags & HAMMER_INODE_RO)
1176                 return (EROFS);
1177         if (tdip->flags & HAMMER_INODE_RO)
1178                 return (EROFS);
1179         if (ip->flags & HAMMER_INODE_RO)
1180                 return (EROFS);
1181
1182         hammer_start_transaction(&trans, fdip->hmp);
1183
1184         /*
1185          * Remove tncp from the target directory and then link ip as
1186          * tncp. XXX pass trans to dounlink
1187          *
1188          * Force the inode sync-time to match the transaction so it is
1189          * in-sync with the creation of the target directory entry.
1190          */
1191         error = hammer_dounlink(ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1192         if (error == 0 || error == ENOENT) {
1193                 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1194                 if (error == 0) {
1195                         ip->ino_data.parent_obj_id = tdip->obj_id;
1196                         hammer_modify_inode(&trans, ip,
1197                                 HAMMER_INODE_DDIRTY | HAMMER_INODE_TIDLOCKED);
1198                 }
1199         }
1200         if (error)
1201                 goto failed; /* XXX */
1202
1203         /*
1204          * Locate the record in the originating directory and remove it.
1205          *
1206          * Calculate the namekey and setup the key range for the scan.  This
1207          * works kinda like a chained hash table where the lower 32 bits
1208          * of the namekey synthesize the chain.
1209          *
1210          * The key range is inclusive of both key_beg and key_end.
1211          */
1212         namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1213 retry:
1214         hammer_init_cursor_hmp(&cursor, &fdip->cache[0], fdip->hmp);
1215         cursor.key_beg.obj_id = fdip->obj_id;
1216         cursor.key_beg.key = namekey;
1217         cursor.key_beg.create_tid = 0;
1218         cursor.key_beg.delete_tid = 0;
1219         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1220         cursor.key_beg.obj_type = 0;
1221
1222         cursor.key_end = cursor.key_beg;
1223         cursor.key_end.key |= 0xFFFFFFFFULL;
1224         cursor.asof = fdip->obj_asof;
1225         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1226
1227         /*
1228          * Scan all matching records (the chain), locate the one matching
1229          * the requested path component.
1230          *
1231          * The hammer_ip_*() functions merge in-memory records with on-disk
1232          * records for the purposes of the search.
1233          */
1234         error = hammer_ip_first(&cursor, fdip);
1235         while (error == 0) {
1236                 if (hammer_ip_resolve_data(&cursor) != 0)
1237                         break;
1238                 rec = cursor.record;
1239                 if (fncp->nc_nlen == rec->entry.base.data_len &&
1240                     bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
1241                         break;
1242                 }
1243                 error = hammer_ip_next(&cursor);
1244         }
1245
1246         /*
1247          * If all is ok we have to get the inode so we can adjust nlinks.
1248          *
1249          * WARNING: hammer_ip_del_directory() may have to terminate the
1250          * cursor to avoid a recursion.  It's ok to call hammer_done_cursor()
1251          * twice.
1252          */
1253         if (error == 0)
1254                 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1255         hammer_done_cursor(&cursor);
1256         if (error == 0)
1257                 cache_rename(ap->a_fnch, ap->a_tnch);
1258         if (error == EDEADLK)
1259                 goto retry;
1260 failed:
1261         if (error == 0) {
1262                 hammer_commit_transaction(&trans);
1263         } else {
1264                 hammer_abort_transaction(&trans);
1265         }
1266         return (error);
1267 }
1268
1269 /*
1270  * hammer_vop_nrmdir { nch, dvp, cred }
1271  */
1272 static
1273 int
1274 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1275 {
1276         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1277 }
1278
1279 /*
1280  * hammer_vop_setattr { vp, vap, cred }
1281  */
1282 static
1283 int
1284 hammer_vop_setattr(struct vop_setattr_args *ap)
1285 {
1286         struct hammer_transaction trans;
1287         struct vattr *vap;
1288         struct hammer_inode *ip;
1289         int modflags;
1290         int error;
1291         int truncating;
1292         int64_t aligned_size;
1293         u_int32_t flags;
1294         uuid_t uuid;
1295
1296         vap = ap->a_vap;
1297         ip = ap->a_vp->v_data;
1298         modflags = 0;
1299
1300         if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1301                 return(EROFS);
1302         if (ip->flags & HAMMER_INODE_RO)
1303                 return (EROFS);
1304
1305         hammer_start_transaction(&trans, ip->hmp);
1306         error = 0;
1307
1308         if (vap->va_flags != VNOVAL) {
1309                 flags = ip->ino_data.uflags;
1310                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1311                                          hammer_to_unix_xid(&ip->ino_data.uid),
1312                                          ap->a_cred);
1313                 if (error == 0) {
1314                         if (ip->ino_data.uflags != flags) {
1315                                 ip->ino_data.uflags = flags;
1316                                 modflags |= HAMMER_INODE_DDIRTY;
1317                         }
1318                         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1319                                 error = 0;
1320                                 goto done;
1321                         }
1322                 }
1323                 goto done;
1324         }
1325         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1326                 error = EPERM;
1327                 goto done;
1328         }
1329         if (vap->va_uid != (uid_t)VNOVAL) {
1330                 hammer_guid_to_uuid(&uuid, vap->va_uid);
1331                 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1332                         ip->ino_data.uid = uuid;
1333                         modflags |= HAMMER_INODE_DDIRTY;
1334                 }
1335         }
1336         if (vap->va_gid != (uid_t)VNOVAL) {
1337                 hammer_guid_to_uuid(&uuid, vap->va_gid);
1338                 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1339                         ip->ino_data.gid = uuid;
1340                         modflags |= HAMMER_INODE_DDIRTY;
1341                 }
1342         }
1343         while (vap->va_size != VNOVAL && ip->ino_rec.ino_size != vap->va_size) {
1344                 switch(ap->a_vp->v_type) {
1345                 case VREG:
1346                         if (vap->va_size == ip->ino_rec.ino_size)
1347                                 break;
1348                         if (vap->va_size < ip->ino_rec.ino_size) {
1349                                 vtruncbuf(ap->a_vp, vap->va_size,
1350                                           HAMMER_BUFSIZE);
1351                                 truncating = 1;
1352                         } else {
1353                                 vnode_pager_setsize(ap->a_vp, vap->va_size);
1354                                 truncating = 0;
1355                         }
1356                         ip->ino_rec.ino_size = vap->va_size;
1357                         modflags |= HAMMER_INODE_RDIRTY;
1358                         aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1359                                         ~(int64_t)HAMMER_BUFMASK;
1360
1361                         if (truncating) {
1362                                 error = hammer_ip_delete_range(&trans, ip,
1363                                                     aligned_size,
1364                                                     0x7FFFFFFFFFFFFFFFLL);
1365                         }
1366                         /*
1367                          * If truncating we have to clean out a portion of
1368                          * the last block on-disk.
1369                          */
1370                         if (truncating && error == 0 &&
1371                             vap->va_size < aligned_size) {
1372                                 struct buf *bp;
1373                                 int offset;
1374
1375                                 offset = vap->va_size & HAMMER_BUFMASK;
1376                                 error = bread(ap->a_vp,
1377                                               aligned_size - HAMMER_BUFSIZE,
1378                                               HAMMER_BUFSIZE, &bp);
1379                                 if (error == 0) {
1380                                         bzero(bp->b_data + offset,
1381                                               HAMMER_BUFSIZE - offset);
1382                                         bdwrite(bp);
1383                                 } else {
1384                                         brelse(bp);
1385                                 }
1386                         }
1387                         break;
1388                 case VDATABASE:
1389                         error = hammer_ip_delete_range(&trans, ip,
1390                                                     vap->va_size,
1391                                                     0x7FFFFFFFFFFFFFFFLL);
1392                         ip->ino_rec.ino_size = vap->va_size;
1393                         modflags |= HAMMER_INODE_RDIRTY;
1394                         break;
1395                 default:
1396                         error = EINVAL;
1397                         goto done;
1398                 }
1399                 break;
1400         }
1401         if (vap->va_atime.tv_sec != VNOVAL) {
1402                 ip->ino_rec.ino_atime =
1403                         hammer_timespec_to_transid(&vap->va_atime);
1404                 modflags |= HAMMER_INODE_ITIMES;
1405         }
1406         if (vap->va_mtime.tv_sec != VNOVAL) {
1407                 ip->ino_rec.ino_mtime =
1408                         hammer_timespec_to_transid(&vap->va_mtime);
1409                 modflags |= HAMMER_INODE_ITIMES;
1410         }
1411         if (vap->va_mode != (mode_t)VNOVAL) {
1412                 if (ip->ino_data.mode != vap->va_mode) {
1413                         ip->ino_data.mode = vap->va_mode;
1414                         modflags |= HAMMER_INODE_DDIRTY;
1415                 }
1416         }
1417 done:
1418         if (error) {
1419                 hammer_abort_transaction(&trans);
1420         } else {
1421                 hammer_modify_inode(&trans, ip, modflags);
1422                 hammer_commit_transaction(&trans);
1423         }
1424         return (error);
1425 }
1426
1427 /*
1428  * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1429  */
1430 static
1431 int
1432 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1433 {
1434         struct hammer_transaction trans;
1435         struct hammer_inode *dip;
1436         struct hammer_inode *nip;
1437         struct nchandle *nch;
1438         hammer_record_t record;
1439         int error;
1440         int bytes;
1441
1442         ap->a_vap->va_type = VLNK;
1443
1444         nch = ap->a_nch;
1445         dip = VTOI(ap->a_dvp);
1446
1447         if (dip->flags & HAMMER_INODE_RO)
1448                 return (EROFS);
1449
1450         /*
1451          * Create a transaction to cover the operations we perform.
1452          */
1453         hammer_start_transaction(&trans, dip->hmp);
1454
1455         /*
1456          * Create a new filesystem object of the requested type.  The
1457          * returned inode will be referenced but not locked.
1458          */
1459
1460         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1461         if (error) {
1462                 hammer_abort_transaction(&trans);
1463                 *ap->a_vpp = NULL;
1464                 return (error);
1465         }
1466
1467         /*
1468          * Add the new filesystem object to the directory.  This will also
1469          * bump the inode's link count.
1470          */
1471         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1472
1473         /*
1474          * Add a record representing the symlink.  symlink stores the link
1475          * as pure data, not a string, and is no \0 terminated.
1476          */
1477         if (error == 0) {
1478                 record = hammer_alloc_mem_record(nip);
1479                 bytes = strlen(ap->a_target);
1480
1481                 record->rec.base.base.key = HAMMER_FIXKEY_SYMLINK;
1482                 record->rec.base.base.rec_type = HAMMER_RECTYPE_FIX;
1483                 record->rec.base.data_len = bytes;
1484                 record->data = (void *)ap->a_target;
1485                 /* will be reallocated by routine below */
1486                 error = hammer_ip_add_record(&trans, record);
1487
1488                 /*
1489                  * Set the file size to the length of the link.
1490                  */
1491                 if (error == 0) {
1492                         nip->ino_rec.ino_size = bytes;
1493                         hammer_modify_inode(&trans, nip, HAMMER_INODE_RDIRTY);
1494                 }
1495         }
1496
1497         /*
1498          * Finish up.
1499          */
1500         if (error) {
1501                 hammer_rel_inode(nip, 0);
1502                 hammer_abort_transaction(&trans);
1503                 *ap->a_vpp = NULL;
1504         } else {
1505                 hammer_commit_transaction(&trans);
1506                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
1507                 hammer_rel_inode(nip, 0);
1508                 if (error == 0) {
1509                         cache_setunresolved(ap->a_nch);
1510                         cache_setvp(ap->a_nch, *ap->a_vpp);
1511                 }
1512         }
1513         return (error);
1514 }
1515
1516 /*
1517  * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1518  */
1519 static
1520 int
1521 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1522 {
1523         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, ap->a_flags));
1524 }
1525
1526 /*
1527  * hammer_vop_ioctl { vp, command, data, fflag, cred }
1528  */
1529 static
1530 int
1531 hammer_vop_ioctl(struct vop_ioctl_args *ap)
1532 {
1533         struct hammer_inode *ip = ap->a_vp->v_data;
1534
1535         return(hammer_ioctl(ip, ap->a_command, ap->a_data,
1536                             ap->a_fflag, ap->a_cred));
1537 }
1538
1539 static
1540 int
1541 hammer_vop_mountctl(struct vop_mountctl_args *ap)
1542 {
1543         struct mount *mp;
1544         int error;
1545
1546         mp = ap->a_head.a_ops->head.vv_mount;
1547
1548         switch(ap->a_op) {
1549         case MOUNTCTL_SET_EXPORT:
1550                 if (ap->a_ctllen != sizeof(struct export_args))
1551                         error = EINVAL;
1552                 error = hammer_vfs_export(mp, ap->a_op,
1553                                       (const struct export_args *)ap->a_ctl);
1554                 break;
1555         default:
1556                 error = journal_mountctl(ap);
1557                 break;
1558         }
1559         return(error);
1560 }
1561
1562 /*
1563  * hammer_vop_strategy { vp, bio }
1564  *
1565  * Strategy call, used for regular file read & write only.  Note that the
1566  * bp may represent a cluster.
1567  *
1568  * To simplify operation and allow better optimizations in the future,
1569  * this code does not make any assumptions with regards to buffer alignment
1570  * or size.
1571  */
1572 static
1573 int
1574 hammer_vop_strategy(struct vop_strategy_args *ap)
1575 {
1576         struct buf *bp;
1577         int error;
1578
1579         bp = ap->a_bio->bio_buf;
1580
1581         switch(bp->b_cmd) {
1582         case BUF_CMD_READ:
1583                 error = hammer_vop_strategy_read(ap);
1584                 break;
1585         case BUF_CMD_WRITE:
1586                 error = hammer_vop_strategy_write(ap);
1587                 break;
1588         default:
1589                 error = EINVAL;
1590                 break;
1591         }
1592         bp->b_error = error;
1593         if (error)
1594                 bp->b_flags |= B_ERROR;
1595         biodone(ap->a_bio);
1596         return (error);
1597 }
1598
1599 /*
1600  * Read from a regular file.  Iterate the related records and fill in the
1601  * BIO/BUF.  Gaps are zero-filled.
1602  *
1603  * The support code in hammer_object.c should be used to deal with mixed
1604  * in-memory and on-disk records.
1605  *
1606  * XXX atime update
1607  */
1608 static
1609 int
1610 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1611 {
1612         struct hammer_inode *ip = ap->a_vp->v_data;
1613         struct hammer_cursor cursor;
1614         hammer_record_ondisk_t rec;
1615         hammer_base_elm_t base;
1616         struct bio *bio;
1617         struct buf *bp;
1618         int64_t rec_offset;
1619         int64_t ran_end;
1620         int64_t tmp64;
1621         int error;
1622         int boff;
1623         int roff;
1624         int n;
1625
1626         bio = ap->a_bio;
1627         bp = bio->bio_buf;
1628
1629         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1630
1631         /*
1632          * Key range (begin and end inclusive) to scan.  Note that the key's
1633          * stored in the actual records represent BASE+LEN, not BASE.  The
1634          * first record containing bio_offset will have a key > bio_offset.
1635          */
1636         cursor.key_beg.obj_id = ip->obj_id;
1637         cursor.key_beg.create_tid = 0;
1638         cursor.key_beg.delete_tid = 0;
1639         cursor.key_beg.obj_type = 0;
1640         cursor.key_beg.key = bio->bio_offset + 1;
1641         cursor.asof = ip->obj_asof;
1642         cursor.flags |= HAMMER_CURSOR_ASOF | HAMMER_CURSOR_DATAEXTOK;
1643
1644         cursor.key_end = cursor.key_beg;
1645         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1646                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1647                 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1648                 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1649         } else {
1650                 ran_end = bio->bio_offset + bp->b_bufsize;
1651                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1652                 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1653                 tmp64 = ran_end + MAXPHYS + 1;  /* work-around GCC-4 bug */
1654                 if (tmp64 < ran_end)
1655                         cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1656                 else
1657                         cursor.key_end.key = ran_end + MAXPHYS + 1;
1658         }
1659         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1660
1661         error = hammer_ip_first(&cursor, ip);
1662         boff = 0;
1663
1664         while (error == 0) {
1665                 error = hammer_ip_resolve_data(&cursor);
1666                 if (error)
1667                         break;
1668                 rec = cursor.record;
1669                 base = &rec->base.base;
1670
1671                 rec_offset = base->key - rec->data.base.data_len;
1672
1673                 /*
1674                  * Calculate the gap, if any, and zero-fill it.
1675                  */
1676                 n = (int)(rec_offset - (bio->bio_offset + boff));
1677                 if (n > 0) {
1678                         if (n > bp->b_bufsize - boff)
1679                                 n = bp->b_bufsize - boff;
1680                         bzero((char *)bp->b_data + boff, n);
1681                         boff += n;
1682                         n = 0;
1683                 }
1684
1685                 /*
1686                  * Calculate the data offset in the record and the number
1687                  * of bytes we can copy.
1688                  *
1689                  * Note there is a degenerate case here where boff may
1690                  * already be at bp->b_bufsize.
1691                  */
1692                 roff = -n;
1693                 n = rec->data.base.data_len - roff;
1694                 KKASSERT(n > 0);
1695                 if (n > bp->b_bufsize - boff)
1696                         n = bp->b_bufsize - boff;
1697                 bcopy((char *)cursor.data + roff,
1698                       (char *)bp->b_data + boff, n);
1699                 boff += n;
1700                 if (boff == bp->b_bufsize)
1701                         break;
1702                 error = hammer_ip_next(&cursor);
1703         }
1704         hammer_done_cursor(&cursor);
1705
1706         /*
1707          * There may have been a gap after the last record
1708          */
1709         if (error == ENOENT)
1710                 error = 0;
1711         if (error == 0 && boff != bp->b_bufsize) {
1712                 KKASSERT(boff < bp->b_bufsize);
1713                 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1714                 /* boff = bp->b_bufsize; */
1715         }
1716         bp->b_resid = 0;
1717         return(error);
1718 }
1719
1720 /*
1721  * Write to a regular file.  Iterate the related records and mark for
1722  * deletion.  If existing edge records (left and right side) overlap our
1723  * write they have to be marked deleted and new records created, usually
1724  * referencing a portion of the original data.  Then add a record to
1725  * represent the buffer.
1726  *
1727  * The support code in hammer_object.c should be used to deal with mixed
1728  * in-memory and on-disk records.
1729  */
1730 static
1731 int
1732 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1733 {
1734         struct hammer_transaction trans;
1735         hammer_inode_t ip;
1736         struct bio *bio;
1737         struct buf *bp;
1738         int error;
1739
1740         bio = ap->a_bio;
1741         bp = bio->bio_buf;
1742         ip = ap->a_vp->v_data;
1743
1744         if (ip->flags & HAMMER_INODE_RO)
1745                 return (EROFS);
1746
1747         /*
1748          * Start a transaction using the TID stored with the bp.
1749          */
1750         KKASSERT(bp->b_tid != 0);
1751         hammer_start_transaction_tid(&trans, ip->hmp, bp->b_tid);
1752
1753         /*
1754          * Delete any records overlapping our range.  This function will
1755          * (eventually) properly truncate partial overlaps.
1756          */
1757         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1758                 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1759                                                bio->bio_offset);
1760         } else {
1761                 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1762                                                bio->bio_offset +
1763                                                 bp->b_bufsize - 1);
1764         }
1765
1766         /*
1767          * Add a single record to cover the write
1768          */
1769         if (error == 0) {
1770                 error = hammer_ip_sync_data(&trans, ip, bio->bio_offset,
1771                                             bp->b_data, bp->b_bufsize);
1772         }
1773
1774         /*
1775          * If an error occured abort the transaction
1776          */
1777         if (error) {
1778                 /* XXX undo deletion */
1779                 hammer_abort_transaction(&trans);
1780                 bp->b_resid = bp->b_bufsize;
1781         } else {
1782                 hammer_commit_transaction(&trans);
1783                 bp->b_resid = 0;
1784                 bp->b_tid = 0;
1785         }
1786         return(error);
1787 }
1788
1789 /*
1790  * dounlink - disconnect a directory entry
1791  *
1792  * XXX whiteout support not really in yet
1793  */
1794 static int
1795 hammer_dounlink(struct nchandle *nch, struct vnode *dvp, struct ucred *cred,
1796                 int flags)
1797 {
1798         struct hammer_transaction trans;
1799         struct namecache *ncp;
1800         hammer_inode_t dip;
1801         hammer_inode_t ip;
1802         hammer_record_ondisk_t rec;
1803         struct hammer_cursor cursor;
1804         int64_t namekey;
1805         int error;
1806
1807         /*
1808          * Calculate the namekey and setup the key range for the scan.  This
1809          * works kinda like a chained hash table where the lower 32 bits
1810          * of the namekey synthesize the chain.
1811          *
1812          * The key range is inclusive of both key_beg and key_end.
1813          */
1814         dip = VTOI(dvp);
1815         ncp = nch->ncp;
1816
1817         if (dip->flags & HAMMER_INODE_RO)
1818                 return (EROFS);
1819
1820         hammer_start_transaction(&trans, dip->hmp);
1821
1822         namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1823 retry:
1824         hammer_init_cursor_hmp(&cursor, &dip->cache[0], dip->hmp);
1825         cursor.key_beg.obj_id = dip->obj_id;
1826         cursor.key_beg.key = namekey;
1827         cursor.key_beg.create_tid = 0;
1828         cursor.key_beg.delete_tid = 0;
1829         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1830         cursor.key_beg.obj_type = 0;
1831
1832         cursor.key_end = cursor.key_beg;
1833         cursor.key_end.key |= 0xFFFFFFFFULL;
1834         cursor.asof = dip->obj_asof;
1835         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1836
1837         /*
1838          * Scan all matching records (the chain), locate the one matching
1839          * the requested path component.  info->last_error contains the
1840          * error code on search termination and could be 0, ENOENT, or
1841          * something else.
1842          *
1843          * The hammer_ip_*() functions merge in-memory records with on-disk
1844          * records for the purposes of the search.
1845          */
1846         error = hammer_ip_first(&cursor, dip);
1847         while (error == 0) {
1848                 error = hammer_ip_resolve_data(&cursor);
1849                 if (error)
1850                         break;
1851                 rec = cursor.record;
1852                 if (ncp->nc_nlen == rec->entry.base.data_len &&
1853                     bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
1854                         break;
1855                 }
1856                 error = hammer_ip_next(&cursor);
1857         }
1858
1859         /*
1860          * If all is ok we have to get the inode so we can adjust nlinks.
1861          *
1862          * If the target is a directory, it must be empty.
1863          */
1864         if (error == 0) {
1865                 ip = hammer_get_inode(dip->hmp, &dip->cache[1],
1866                                       rec->entry.obj_id,
1867                                       dip->hmp->asof, 0, &error);
1868                 if (error == ENOENT) {
1869                         kprintf("obj_id %016llx\n", rec->entry.obj_id);
1870                         Debugger("ENOENT unlinking object that should exist, cont to sync");
1871                         hammer_sync_hmp(dip->hmp, MNT_NOWAIT);
1872                         Debugger("ENOENT - sync done");
1873                 }
1874                 if (error == 0 && ip->ino_rec.base.base.obj_type ==
1875                                   HAMMER_OBJTYPE_DIRECTORY) {
1876                         error = hammer_ip_check_directory_empty(&trans, ip);
1877                 }
1878                 /*
1879                  * WARNING: hammer_ip_del_directory() may have to terminate
1880                  * the cursor to avoid a lock recursion.  It's ok to call
1881                  * hammer_done_cursor() twice.
1882                  */
1883                 if (error == 0)
1884                         error = hammer_ip_del_directory(&trans, &cursor, dip, ip);
1885                 if (error == 0) {
1886                         cache_setunresolved(nch);
1887                         cache_setvp(nch, NULL);
1888                         /* XXX locking */
1889                         if (ip->vp)
1890                                 cache_inval_vp(ip->vp, CINV_DESTROY);
1891                 }
1892                 hammer_rel_inode(ip, 0);
1893         }
1894         hammer_done_cursor(&cursor);
1895         if (error == EDEADLK)
1896                 goto retry;
1897
1898         if (error == 0)
1899                 hammer_commit_transaction(&trans);
1900         else
1901                 hammer_abort_transaction(&trans);
1902         return (error);
1903 }
1904
1905 /************************************************************************
1906  *                          FIFO AND SPECFS OPS                         *
1907  ************************************************************************
1908  *
1909  */
1910
1911 static int
1912 hammer_vop_fifoclose (struct vop_close_args *ap)
1913 {
1914         /* XXX update itimes */
1915         return (VOCALL(&fifo_vnode_vops, &ap->a_head));
1916 }
1917
1918 static int
1919 hammer_vop_fiforead (struct vop_read_args *ap)
1920 {
1921         int error;
1922
1923         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1924         /* XXX update access time */
1925         return (error);
1926 }
1927
1928 static int
1929 hammer_vop_fifowrite (struct vop_write_args *ap)
1930 {
1931         int error;
1932
1933         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1934         /* XXX update access time */
1935         return (error);
1936 }
1937
1938 static int
1939 hammer_vop_specclose (struct vop_close_args *ap)
1940 {
1941         /* XXX update itimes */
1942         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1943 }
1944
1945 static int
1946 hammer_vop_specread (struct vop_read_args *ap)
1947 {
1948         /* XXX update access time */
1949         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1950 }
1951
1952 static int
1953 hammer_vop_specwrite (struct vop_write_args *ap)
1954 {
1955         /* XXX update last change time */
1956         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1957 }
1958